🎯 Intelligent Decisions for Logistics

An Introduction to Bayesian Optimization

Witek ten Hove

Welcome

Target Audience: Business Professionals in Logistics
Duration: 30 Minutes
Interactive Experience: Click, Explore, Learn

Welcome to Smart Optimization for the Real World

🎯 Your Learning Journey

By the end of this session, you will:

🎯 Identify Opportunities

Spot areas where traditional ‘trial and error’ is too slow or expensive

🚀 Get Better Results, Faster

Understand how smart optimization finds solutions with fewer tests

🗣️ Speak the Language

Understand ‘smart predictors’ and ‘decision-makers’ for data science conversations

⚖️ Balanced Strategy

Grasp why exploration vs exploitation balance drives innovation

Interactive demos ahead - get ready to click and explore!

📋 Today’s Roadmap

🎯

The Challenge
2 minutes

🧠

Smart Strategy
7 minutes

🔄

The Loop
4 minutes

🚛

Real Impact
2 minutes

🤖 The Problem We’re Solving

In logistics, every test costs time, energy, and money.
How do we find the best solutions with the fewest expensive trials?

🎯 The Challenge: Expensive Black-Box Problems

🔲 What is a “Black-Box”?

Think: New route planning software

  • You enter settings (capacity, delivery windows)
  • It gives you an efficiency score
  • You can’t see the complex calculations inside
  • You only know: INPUT → OUTPUT

💰 Why is Testing “Expensive”?

Every evaluation costs:

  • Time: Weeks for pilot tests
  • Money: Fuel, wages, disruptions
  • Resources: Vehicles, staff, computing

❌ You can’t run millions of tests!

⚡ The Dilemma

⚡ The Dilemma

You want the best result, but testing every combination is impossible.
Random guesses are inefficient. We need a smarter strategy.

Next: How Bayesian Optimization solves this intelligently →

🧠 The Smart Strategy: Bayesian Optimization

🎭 Two Smart Components Working Together

Like having an intelligent assistant that gets smarter with every test

🔮 “Smart Predictor”

(Surrogate Model)

Learns from your test results to predict what might happen everywhere else

Like an experienced manager predicting route performance

🎯 “Smart Decision Maker”

(Acquisition Function)

Decides where to run your next expensive test for maximum value

Balances exploring new areas vs improving known good areas

🎯 Bayesian Optimization Learning Process

From Uncertainty to Confidence: The Learning Journey

Watch how the Smart Predictor becomes more accurate with each test

Figure 1

🎯 How the “Smart Decision Maker” Finds the Optimum

Exploit vs. Explore

Chooses the next test to either improve the best-known result or reduce uncertainty in a new area.

Figure 2

🔄 The Bayesian Optimization Loop

5 Simple Steps to Smart Optimization

1

🎯 Start

Run initial tests

2

🔮 Predict

Update belief

3

🎯 Decide

Next test location

4

🧪 Test

Run experiment

5

🔄 Repeat

Learn & iterate

💡 Key Insight

Notice how the algorithm first explores uncertain areas, then exploits promising regions to find the optimum faster than random testing!

🏭 See It In Action: Finding the Optimal DC Location

Challenge: Find optimal DC location for 30 customers with minimal $50K feasibility studies

3D Bayesian Optimization Visualization

🚀 Launch Simulator
Opens in new window

🔍 What You’ll See

Blue cylinders: Customer locations
Colored surface: AI’s belief about costs
Red wireframe: True hidden cost landscape
Purple surface: Next test decision guide

⚡ Try These

• Click “Initialize/Reset” to start
“Run One Step” for learning
• Adjust Exploration (Beta) slider
• Toggle surfaces to see beliefs vs reality
• Watch it find the hidden optimum!

💡 Key Discovery: Watch how the algorithm intelligently balances exploring uncertain areas with exploiting promising locations to find the optimal solution.

🚛 Real-World Impact in Logistics

Your Competitive Advantage

Reducing expensive trials = Efficiency + Cost Savings + Innovation

🤖

ML Model Optimization

Find the best settings for demand forecasting and delivery time estimation models

🚚

Route Optimization

Fine-tune vehicle routing algorithms to minimize fuel costs and delivery times

🏭

Warehouse Robotics

Optimize robot movement and sorting policies using efficient simulations

⛓️

Supply Chain Design

Find optimal distribution center locations balancing costs and service levels

📦 The Bottom Line

When each test costs time and money,
smart optimization isn’t just nice to have—it’s essential for staying competitive

🚀 The Next Frontier: AI Discovering New Strategies

AI Teaching AI: The Future of Optimization

AI now creates entirely new optimization approaches

🤖 Meet FunBO: AI Creating Smart Strategies

🧠

AI Writes Code: LLM generates new decision-making strategies

🧪

Auto-Testing: System tests each strategy on different problems

🏆

Evolution: Best strategies survive and improve

💡 Business Impact

Custom strategies that work better than standard approaches

Source: Aglietti et al. (2024). Funbo: Discovering acquisition functions for bayesian optimization with funsearch. arXiv:2406.04824.

FunBO Process Diagram

How FunBO Works: AI proposes code, tests it, evolves solutions

🎯 Key Takeaways & Next Steps

🧠 What You’ve Learned

Bayesian Optimization = Smart Predictions + Intelligent Decisions

🎉 Key Takeaways

  • Smart Strategy: Smart predictions + decisions
  • Efficient Testing: Fewer trials needed
  • Balanced Approach: Explore + exploit balance
  • Real Impact: Cost & efficiency gains

🚀 Your Next Steps

  1. Identify Opportunities: Find expensive trials
  2. Engage Teams: Talk to data teams
  3. Start Small: Pilot one problem
  4. Measure Results: Track results

🎤 Questions & Discussion

🎤

Ready to explore how Bayesian Optimization can transform your logistics operations?
Let’s discuss your specific challenges and opportunities!

🙏 Thank You!

Contact Information
HAN Lectoraat Logistiek en Allianties | Karen.Engelvaart@han.nl

Generated with Bayesian Optimization interactive demos